Levidow LeGendre AI Citation Hallucination: Court Sanctions Incident Record

Incident Summary

In 2023, attorneys at the New York personal injury firm Levidow, Levidow & Oberman (also identified in some reports as Levidow LeGendre) submitted a legal brief in a personal injury matter in New York state court that contained multiple case citations generated by an AI tool — citations that, upon judicial review, did not correspond to any real decisions. The court found that the attorneys had relied on ChatGPT to research and draft portions of the filing without independently verifying the cited authorities.

When the opposing party and the court could not locate the cited cases, the attorneys were ordered to provide copies of the decisions. They could not produce them. Subsequent investigation confirmed that the citations — including case names, reporters, and page numbers — were entirely fabricated by the AI system.

Structured Record

What Happened: The Filing and Its Defects

The attorneys used ChatGPT to identify case law supporting their client's position in a personal injury action. The AI returned a set of case citations — complete with party names, court identifiers, year references, and reporter citations — that appeared, on their face, to be legitimate legal authority.

None of those cases existed. The AI had generated plausible-sounding but entirely fictional citations. This is a well-documented failure mode of large language models: they produce text that conforms to the structural conventions of legal citations without having any grounding in actual case databases.

The attorneys did not run the citations through Westlaw, Lexis, or any other primary legal research database before submitting the brief. When the court and opposing counsel flagged the unlocatable cases, the attorneys initially represented that the cases did exist — compounding the original error with a secondary misrepresentation to the court.

Sanctions and Court Orders

The presiding judge imposed monetary sanctions on the attorneys and required them to submit affidavits in future filings attesting that cited authorities had been independently verified against primary legal databases. The court's written order addressed the use of generative AI in legal practice directly, noting that AI tools do not access live legal databases and are capable of fabricating citations that appear superficially authentic.

The sanction amount was not publicly disclosed in all reporting, but the court order itself — and the requirement for ongoing affidavits — created a compliance burden on the firm that extends beyond the single matter.

Context: A Pattern, Not an Isolated Event

The Levidow LeGendre incident occurred in the same period as the higher-profile Mata v. Avianca sanctions matter in the Southern District of New York, where attorneys Steven Schwartz and Peter LoDuca at Levin & Associates also submitted ChatGPT-generated fictitious citations. The Mata matter received more press coverage, but the Levidow LeGendre incident represents a separate, independent occurrence of the same failure mode — demonstrating that citation hallucination was not a one-off event but a recurring problem across different firms and courts in 2023.

Both incidents share an identical structural cause: attorneys used a general-purpose LLM for legal research without understanding that the model generates text probabilistically and has no connection to verified legal databases. ChatGPT, at the time of these filings, had no live Westlaw or Lexis integration in its base form — it could only produce text that resembled legal citations based on patterns in its training data.

Professional Responsibility Implications

Competence Under Model Rule 1.1

ABA Model Rule 1.1 requires attorneys to maintain competence, including — per Comment 8, added in 2012 — the "benefits and risks associated with relevant technology." The Levidow LeGendre incident is routinely cited in bar ethics opinions and CLE materials as a concrete illustration of what Rule 1.1 technology competence failures look like in practice: not a data breach or a confidentiality issue, but a failure to understand what an AI tool actually does and does not do.

Candor Obligations Under Model Rule 3.3

Model Rule 3.3 prohibits making false statements of law or fact to a tribunal and requires attorneys to correct material misstatements. In this incident, the attorneys' initial response — defending the citations as real — is the conduct that most clearly implicates Rule 3.3. The original submission error might have been treated more leniently had the attorneys immediately acknowledged the AI-generated nature of the citations upon challenge.

Supervisory Responsibility Under Model Rule 5.1

Where supervising attorneys direct or permit associates to use AI tools for research without establishing verification protocols, Model Rule 5.1 (Responsibilities of Partners, Managers, and Supervisory Lawyers) becomes relevant. The Levidow LeGendre matter involved attorneys at different seniority levels, raising questions about whether adequate supervision was in place for AI-assisted work product.

What the Court Record Establishes

• The attorneys affirmatively disclosed using ChatGPT for legal research after the citation errors were identified.
• No independent verification against Westlaw, Lexis, or any other primary database was performed before the brief was filed.
• The AI system was not a legal-specific research tool — it was a general-purpose LLM without live database access.
• The court's sanctions order explicitly addressed the limitations of AI tools and the attorney's duty to verify.
• Ongoing compliance obligations (affidavit requirements) were imposed as a forward-looking remedy, not merely a punitive one.

Verification Failure: The Specific Gap

The failure here was not that AI was used — it was that AI output was treated as a terminal research product rather than a starting point requiring verification. Legal research tools that are specifically built for attorney use (such as Westlaw CoCounsel or Lexis+ AI) retrieve citations from verified, curated legal databases and can cite to primary sources. General-purpose LLMs do not.

Downstream Impact on Court Rules and Bar Guidance

The Levidow LeGendre incident, along with Mata v. Avianca, prompted a wave of court-specific AI disclosure rules in 2023 and 2024. Several federal district courts — including the Northern District of Texas, the District of New Mexico, and others — adopted standing orders requiring attorneys to certify that AI-generated content had been reviewed for accuracy and that citations had been independently verified.

State bar associations also responded. Multiple bars issued formal ethics opinions in 2023–2024 addressing attorney AI use, with citation verification consistently identified as a non-delegable professional responsibility obligation. The incident record for Levidow LeGendre thus functions as a traceable origin point for a significant body of subsequent regulatory guidance.